Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A/B test of banner ads vs. traffic

gwern.net

Post reply on HN
Print view View original

141–149 of 149 posts

Re: A/B test of banner ads vs. traffic

#141

Post by KajMagnus »

pault wrote: post #63
magicmikexxl wrote: post #9

There are ways to have comments that are a lot less awful than Disqus.

I don't like Disqus mostly because the UX is clunky. Are there other things people dislike about it?

Disqus shows ads, unless you pay like $10/month. (& tracks your visitors, like someone else mentioned.)

Maybe you would instead like the looks of EffectiveDiscussions — it's a new commenting system I'm developing. See my other comment nearby (or, if you didn't find it, go to: https://www.effectivediscussions.org/blog-comments ).

Re: A/B test of banner ads vs. traffic

#142

Post by spc476 »

yeldarb wrote: post #126
spc476 wrote: post #118

Earlier quoted context omitted.

If it was a banner of products related to previous purchases, okay, I can see that. But the banner was just a generic "Shop at Amazon!" Lame.

Do you use Ghostery or other “do not track” tools? It’s possible your users were seeing something different

Nope. I have no ad blockers installed. I also checked the Amazon affiliate website and the new banner ads (at the size I wanted) no longer have the option I was used to.

Re: A/B test of banner ads vs. traffic

#143

Post by viraptor »

dredmorbius wrote: post #100
viraptor wrote: post #88

Earlier quoted context omitted.

Something's not right with that equation. You're implying that on average, each household is spending over $1000 per month on things they get via the online advertising that they would otherwise not get at all. And that's just not realistic. One-off purchases, maybe. Sustained 1k/hh/mth avg - I'd love to see a proof of that.

Total advertising spend worldwide is $600 billion. Online is about 1/6 of that, call it $100 billion. To a rough approximation, that's footed by the 1 billion wealthy inhabitants of the EU, US, Japan, Canada, Australian, and New Zealand. (I said rough , roll with me.) That means that the per person costs of advertising are pretty much $100 annually online and $600 annually for all content (mostly TV). Keep in mind, t…

I find it very disappointing and would like to see the ad industry die in a fire :-(

Re: A/B test of banner ads vs. traffic

#144

Post by jackgolding »

gwern wrote: post #56
portman wrote: post #46

Earlier quoted context omitted.

Isn't this particular test susceptible to confounding effects? Traffic fluctuates day-to-day, week-to-week, and month-to-month, so how can you be sure it was the presence-or-absence of ads and not something else? If you randomize at the visitor level, you are sampling from both high-and-low traffic days, and control for any external fluctuations.

Because you're randomizing at the 2-day level, on average there will be just as many advertising/high-traffic days as advertising/low-traffic days, and as many no-advertising/high-traffic days as no-advertising/low-traffic days. The randomization is unaffected by traffic and uncorrelated with it. The unit of analysis is each day, not each visitor. This is why it has to be run for several months, otherwise you don't w…

The stats went well over my head but as a web analyst I thought maybe you could have asked a simpler question. Pick a segment of your site such as visits from Google Search who landed on your homepage (most likely people who searched "gwern") which should reduce a lot of those spikes.

Re: A/B test of banner ads vs. traffic

#145

Post by _8j50 »

bertil wrote: post #69
_8j50 wrote: post #26

Earlier quoted context omitted.

As a user,I am most offended by ads that target me. I would rather see a completely unrelatable ad. It feels like learning new information that way,as opposed to being stalked and preyed upon.

I do not know you, so I won’t presume (certainly not on HackerNews, the least representative sample of the Internet) but I have seen many experiments proving that most people who think they don’t like targetted ads actually prefer them -- until they make a very informed but very wrong guess. Typical examples include: - advertising for something shameful (underwear and dildo are my go-to examples) or that you otherwis…

Not true,sure I would prefer programming related ads but I still want it page-relevant and not based on my browsing history,habits or sensor data.

Re: A/B test of banner ads vs. traffic

#146

Post by gwern »

jackgolding wrote: post #144
gwern wrote: post #56

Earlier quoted context omitted.

Because you're randomizing at the 2-day level, on average there will be just as many advertising/high-traffic days as advertising/low-traffic days, and as many no-advertising/high-traffic days as no-advertising/low-traffic days. The randomization is unaffected by traffic and uncorrelated with it. The unit of analysis is each day, not each visitor. This is why it has to be run for several months, otherwise you don't w…

The stats went well over my head but as a web analyst I thought maybe you could have asked a simpler question. Pick a segment of your site such as visits from Google Search who landed on your homepage (most likely people who searched "gwern") which should reduce a lot of those spikes.

Subsetting will also increase the variance of each datapoint (consider the extreme case of picking a subset which was 0 or 1 visits per day), so is probably not a win. It's also hard to imagine what subset properly reflects all sources of traffic and so is informative about the total effect of advertising. Search queries definitely is not it.

Re: A/B test of banner ads vs. traffic

#147

Post by spiderfarmer »

burntwater wrote: post #77
spiderfarmer wrote: post #45

Earlier quoted context omitted.

I scanned the article to see if it was interesting enough to dive into the details, but when I saw the Disqus plugin I closed the tab. What's the point of dropping Adsense if you keep the Disqus plugin?

Simple, they value the discussion capabilities provided by Disqus. Disqus was worth keeping, Adsense was not.

That’s my point exactly. Did they do the same exercise for Disqus? Or did they get all scientific on evaluating Adsense, and go with a “hunch” for Disqus?

That’s like saying “Marlboro cigarettes are harmful, here’s proof, let’s grab a pack of Lucky Strikes and discuss”.

Re: A/B test of banner ads vs. traffic

#148

Post by FridgeSeal »

techdragon wrote: post #116
FridgeSeal wrote: post #81

Earlier quoted context omitted.

As someone who used to work trafficking ads and spends a lot of time blocking them myself, I'm of the opinion that this is the approach to take. ReadTheDocs and sites like BikeRadar are my favourite examples: If I go to a bike site, I expect the ads I see to be about bike stuff. If I'm on a technical site (like Stack Overflow or ReadTheDocs) I expect to see ads about related technical topics. Unfortunately the advert…

We could build these things, but part of the "advertising host" optimisation metrics is "time to ad display" so the algorithms in use are frequently "deliberately dumb" for reasons of optimising the time involved in deciding what advertising should be displayed.

Web pages are far more static than users, your adserver/ML/analysis tool need only analyse the page once (not including updates for content changes) and then pass this information straight on to ad buyers.

Even if advertisers analysed the page content on seeing it, rather than the publisher, it would still be net faster than doing lookups/analysis per adserve (as you would when you target a user).

I don't think it would be hard at all to build something which analysed a page and made an adserving decision with a lower "time to ad display" than a similar process that used user data.

Re: A/B test of banner ads vs. traffic

#149

Post by gwern »

spiderfarmer wrote: post #147
burntwater wrote: post #77

Earlier quoted context omitted.

Simple, they value the discussion capabilities provided by Disqus. Disqus was worth keeping, Adsense was not.

That’s my point exactly. Did they do the same exercise for Disqus? Or did they get all scientific on evaluating Adsense, and go with a “hunch” for Disqus? That’s like saying “Marlboro cigarettes are harmful, here’s proof, let’s grab a pack of Lucky Strikes and discuss”.

It's more like smoking two brands of cigarettes and then deciding to stop half your smoking; that is surely an improvement...

This is just whataboutism. AdSense should be judged on its own merits, and it fails badly: it does nothing I want (aside from delivering quite small amounts of money) at an unacceptable cost. Maybe Disqus is also bad but I would have to ponder how much I want commenting and how to deal with the extant corpus of comments and in any case, at the moment I am gearing up for another ad test to try to figure out what about AdSense is so harmful (and help convince the doubters that there is an effect at all), so I can't test Disqus without complicating things a lot.

Post reply on HN